I'm running on Ubuntu system with TensorFlow-GPU version 1.13 and CUDA 10.0. The loss becomes NaN from the first epoch. I've tried many methods but none of them worked. Have you encountered this issue? What's your environment like when you run TensorFlow?
